THE ARTS
CHILD POLICY
CIVIL JUSTICE
EDUCATION
ENERGY AND ENVIRONMENT
This PDF document was made available from www.rand.org as a public
service of the RAND Corporation.
Jump down to document6
HEALTH AND HEALTH CARE
INTERNATIONAL AFFAIRS
NATIONAL SECURITY
POPULATION AND AGING
PUBLIC SAFETY
SCIENCE AND TECHNOLOGY
SUBSTANCE ABUSE
TERRORISM AND
HOMELAND SECURITY
TRANSPORTATION AND
INFRASTRUCTURE
WORKFORCE AND WORKPLACE
The RAND Corporation is a nonprofit research
organization providing objective analysis and effective
solutions that address the challenges facing the public
and private sectors around the world.
Support RAND
Browse Books & Publications
Make a charitable contribution
For More Information
Visit RAND at www.rand.org
Explore RAND Health
View document details
This product is part of the RAND Corporation reprint series. RAND reprints
present previously published journal articles, book chapters, and reports with
the permission of the publisher. RAND reprints have been formally reviewed
in accordance with the publisher’s editorial policy, and are compliant with
RAND’s rigorous quality assurance standards for quality and objectivity.
M. Susan Marquis
Stephen H. Long
Effects of ‘‘Second
Generation’’ Small Group
Health Insurance Market
Reforms, 1993 to 1997
In the mid-1990s, several state legislatures enacted a ‘‘second generation’’ of small group
health insurance reforms that required guaranteed issue of all products and prohibited
the use of health as a rating factor. We use data from two large employer surveys to
compare the behavior of small business in nine states that adopted these reforms between
1993 and 1997 to the behavior of small business in 11 states and the District of
Columbia, where neither of these small group health insurance market reforms existed
prior to 1997 (N ⫽ 8,465 in 1993; N ⫽ 12,219 in 1997). Our analyses focus on several
outcomes: health insurance offer and enrollment rates in any employer plan, and in an
HMO plan; turnover in offer decisions; and premiums, variability in premiums, and the
rate of change in premiums. Overall, we Ànd no effect of small group reform on any of
the outcomes; the sign of the effect is not consistent across reform states, the estimates
rarely attain statistical signiÀcance, and they show no consistent pattern across the
outcomes within each state. Therefore, predictions of the harm these regulations might
cause to the market have not come to pass. On the other hand, proponents’ hopes for a
solution to low coverage rates among small businesses have not materialized either.
Many states enacted insurance market reforms during
the 1990s to try to improve access to and affordability of health insurance coverage for employees in
small Àrms. These reforms governed rules of issue
and how insurers could set premiums for coverage.
Reforms related to rules of issue were intended to
eliminate insurer underwriting practices that prevented groups or some individuals in a group from purchasing coverage. Such reforms included: requirements that insurers issue a plan to all employers or
individuals who wish to purchase it (guaranteed issue); requirements that coverage be renewed (guaranteed renewal); limits on the period during which
pre-existing conditions can be excluded from coverage; and exemptions from pre-existing condition ex-
clusions for people who change group insurers (portability). By 1997, 47 states had enacted some combination of access reforms (Curtis et al. 1999). Following these independent state efforts, the federal
government mandated comprehensive access reforms
for all states by enacting the Health Insurance Portability and Accountability Act of 1996 (HIPAA).
Rating reforms restrict the extent to which premiums for a given set of beneÀts can vary across groups
with different characteristics. Rating reforms are intended to enlarge the risk pool on which premiums for
small employers are based, thus making insurance
more affordable for high-risk groups and encouraging
them to purchase insurance. On the other hand, small
employers that have a good risk proÀle may pay high-
M. Susan Marquis, Ph.D., and Stephen H. Long, Ph.D., are senior economists at RAND. This research was supported by
the Robert Wood Johnson Foundation (grants no. 026935 and 028651) and the National Center for Health Statistics (contract
no. 0009930281). Address correspondence to the authors at RAND, 1200 S. Hayes St., Arlington, VA 22202-5050.
Inquiry 38: 365–380 (Winter 2001/2002). 䉷 2001 Blue Cross and Blue Shield Association and
Blue Cross and Blue Shield of the Rochester Area.
0046-9580/01/3804–0365$1.25
365
Inquiry/Volume 38, Winter 2001/2002
er rates after reforms are enacted because insurers will
be limited in the extent to which they can take into
account this favorable proÀle. The federal government
did not address rate regulation under HIPAA; hence,
this type of reform is solely a state option.
The effect that insurance reforms have on the small
group market is theoretically ambiguous. If underwriting restrictions and rating reforms attract high-risk
groups into the market, then average premiums may
rise and some businesses may drop coverage. Some
opponents of these reforms fear that they will drive
the small group market into a ‘‘death spiral,’’ thereby
contributing to rising rates of uninsured (see for example Curtis et al. 1999; Blumberg and Nichols 1996;
Thorpe 1992). However, proponents point out that reforms may encourage greater price competition by
limiting the extent to which insurers can compete on
the basis of risk selection (Buchmueller and Jensen
1997). This may lead to lower prices and more coverage.
In this paper, we examine the effect of guaranteed
issue provisions and rating reforms enacted in the mid1990s on the extent and nature of employer-sponsored
insurance for small groups. We extend the previous
literature by focusing on a time during which a number of states moved to tighten rating reforms and to
require guaranteed issue of all insurance products. We
also examine a wider range of potential outcomes of
reforms than most of the earlier literature. The continuing problem of the uninsured has led to new proposals to try to combat it. Some call for greater regulation of insurance markets (Herzlinger 2000). Others
call for subsidies to private coverage, but observe that
the success of these efforts will depend on having
well-functioning insurance markets (Bilheimer and
Colby 2001). Thus, understanding the role that market
reforms have played is important to the design of future policy intended to increase the proportion of insured workers in small Àrms.
Literature Review
The empirical evidence on the effect of small group
insurance market reforms is somewhat ambiguous.
Early case studies of reforms in speciÀc states found
little change in coverage overall, though some states
experienced improvements in coverage while others
experienced a loss (Chollet and Paul 1995; Wessner
1994; Institute for Health Policy Solutions 1995).
More recent analyses have used primarily two
types of data. One set of studies has used data from
the Current Population Survey (CPS) to investigate
how the number of uninsured and the number cov366
ered by private insurance of all types varies as market
reforms vary across states or time or, in most cases,
both. Most of these studies investigated changes in
the uninsured rate or private coverage rate either for
the entire nonelderly population or for all adults. This
potentially understates the effect of reforms on the
smaller target population of employees in small business. Custer (1998), using data from the 1998 CPS,
concluded that both guaranteed issue and rating reforms led to increases in the number of uninsured.
However, this conclusion was based on differences
in uninsured rates in a cross-section of states and may
reÁect geographic differences other than market regulation. Several analysts have pooled data from multiple years of the CPS to control for state-speciÀc
effects in the uninsured rate in analyzing the role of
reforms. Two of these studies, one for 1989–94 and
a second for 1989–95, found that neither access nor
rating reforms in the small group market affected the
uninsured rate (Sloan and Conover 1998; Zuckerman
and Rajan 1999).1 In contrast, a third study using data
from 1989–95 concluded that guaranteed issue regulations and rating reforms had statistically signiÀcant but countervailing effects on the uninsured rate
(Marsteller et al. 1998).2 Guaranteed issue was associated with a small decrease of about 1 percentage
point in the uninsured rate, which was offset when
rating reforms were also in place. However, Buchmueller and DiNardo (1998) concluded that strict
community rating reforms do not affect uninsured
rates among all adults or among workers in small
Àrms; their conclusion was based on a comparison of
trends in CPS uninsured rates in a state enacting community rating (New York), with trends in a state with
no reforms (Pennsylvania) and a state with only modest reforms (Connecticut).
Another strand of literature has analyzed employer
data to investigate the effect of insurance market reforms on employer decisions to offer insurance. In
contrast to the population analyses, the employer-level analyses focus on the target population of small
business. These studies, too, have produced mixed
results. Hing and Jensen (1999) used variation among
states in the insurance market reforms that had been
enacted prior to 1993 to study access and rating reforms. They concluded that the probability of a small
Àrm offering insurance was about two to three percentage points higher in states that had reforms in
place for several years than in states that had not
enacted reforms by 1993. As the authors observe,
however, states that had better functioning markets
may have been states to enact reforms early, and this
‘‘Second Generation’’ Reforms
self-selection may account for their cross-section results. Jensen and Morrisey (1999) pooled time-series
and cross-sectional data on small employers to analyze reforms enacted over the period 1989 to 1995.
Overall, they did not Ànd statistically signiÀcant effects of access reforms. However, sample size—and
hence power—was small and point estimates suggested that reforms increased small Àrm offer rates
by about two to seven percentage points. Similarly,
Simon (1999) reports small and generally insigniÀcant effects of reforms enacted between 1993 and
1997 on employer offer rates.3 Buchmueller and Jensen (1997) also report that more small Àrms in California offered health insurance following comprehensive reforms that went into effect in 1993; this
result was especially pronounced among Àrms with
fewer than 10 employees. However, similar patterns
were not found in other states enacting reforms during the study period. Moreover, an increase in offer
rates of similar magnitude was seen among very
small Àrms in states that had enacted reforms prior
to the study period, suggesting temporal trends might
be at work.
In sum, the preponderance of evidence suggests
that small group market reforms have had little effect
on the decisions of small employers to offer insurance. However, many of these previous studies
looked at reforms enacted in the early ’90s. Subsequently, a number of states moved to tighter issue
and rating reforms, which might have a greater effect
than the ‘‘Àrst generation’’ of reforms. In addition,
most of the earlier studies did not differentiate the
extent of reform—for example, guaranteed issue for
all products vs. only some standard products, or tight
vs. loose rating restrictions. As a consequence, they
might underestimate the effect of stringent reform. A
contribution of our research is that we used a detailed
database describing small group insurance market
regulations in each state during the period 1990 to
1997 to differentiate the extent of reform.4 Also, we
studied reforms adopted between 1994 and 1996,
which was a period when states were moving toward
tighter rating regulations—especially prohibition of
the use of health as a rating factor. In contrast, rating
reforms covered in many of the earlier studies permitted the use of health ratings, though often limiting
the permissible range of variation associated with
health differences among groups (Curtis et al. 1999).
Finally, we go beyond an exploration of the effect of
the reforms on employer offer rates and employee
enrollments to examine a number of other hypotheses
about reforms, as described later.
Data and Methods
Data
Our data are from the National Employer Health Insurance Survey (NEHIS) and the 1997 Robert Wood
Johnson Foundation (RWJF) Employer Health Insurance Survey. The NEHIS, sponsored by the National
Center for Health Statistics, completed interviews
with 34,604 private employers with one or more employees in 1994. The nationwide sample was designed to provide reliable state-level estimates of employers and employees.5 The 1997 RWJF survey,
conducted by RAND and the Research Triangle Institute, interviewed 21,545 private employers in the
continental United States. The sample was concentrated in the 60 communities followed by the RWJF
Community Tracking Study and in states having signiÀcant small group rating reforms. These cases were
supplemented by a sample from the remainder of the
continental United States to better represent the nation’s business establishments.6
The sampling frame for both surveys was Dun’s
Market IdentiÀers national census of employment establishments. The NEHIS sample was stratiÀed by
state, and within state by Àrm and establishment size.
(Firm size refers to the number of employees at all
locations of a business nationwide; establishment size
refers to employment at a single location of a business.) The RWJF survey was stratiÀed by geographic
area (as deÀned previously) and by size of the establishment. Sampled establishments in each study are
weighted to account for different sampling probabilities and for nonresponse. The weighted samples in
both surveys represent all private employment establishments that have at least one employee.7 The NEHIS survey included establishments in all 50 states and
the District of Columbia, whereas the RWJF covered
establishments in the continental United States. We restricted our analysis of the NEHIS to the geography
covered by the RWJF survey. The distribution of
weighted establishments by establishment size, Àrm
size, and industry was similar across the surveys. In
addition, as we describe next, our analysis controls for
employer characteristics and, hence, for any small differences between the realized samples.
The surveys used similar questions and deÀnitions.
Both collected information from employers using
computer-assisted telephone interviews (CATI). We
subjected the databases to the same algorithms to edit
data for consistency and outlier values, and to impute
missing data. Both surveys were conducted with the
person or persons in each establishment most knowl367
Inquiry/Volume 38, Winter 2001/2002
edgeable about health beneÀts and business and worker characteristics. The NEHIS survey was administered during 1994 and asked respondents to report
characteristics of coverage as of the end of 1993. The
RWJF survey was administered during 1997 and employers were asked about coverage as of the date of
the interview. The response rate to the NEHIS was
71%; for the RWJF survey, it was 60%. These are
comparable to response rates for two other recent
large-scale surveys of employer health insurance (Gabel, Ginsburg, and Hunt 1997; Branscome et al. 2000).
The surveys collected information about the employer, the number of workers, and the distribution
of key characteristics of these workers. They also collected detailed information about every health insurance plan offered by the business, including the number of employees enrolled in the plan, premiums, and
the type of plan. In addition, the RWJF survey asked
about the percentage change in premiums across all
offered plans that each employer experienced between 1996 and 1997. The establishment is the unit
of observation for most analyses in this paper. However, the numbers of employees and health plan enrollees at each establishment also allow us to develop
sampling weights to make estimates for employees
and enrollees for some analyses.
Estimation Strategy
Our analysis of the effects of reforms focuses on
guaranteed issue requirements and rating reforms that
prohibit the use of health status as a factor. We do
this because these represent the ‘‘second generation’’
of reforms—few states required guaranteed issue of
all products nor prohibited rating based on health status prior to 1993.8 We examine nine states that enacted these reforms between 1993 and 1996. Thus,
the NEHIS data from 1993 provide outcomes prior
to reforms, and the RWJF survey data from 1997
provide outcomes subsequent to reforms.9 Six of the
reform study states had neither guaranteed issue nor
health rating prohibitions prior to 1993 but added one
of these provisions to law between 1993 and 1996.
Three of the states required that some products be
guaranteed issue prior to 1993, but enacted tighter
issue regulations or health rating prohibitions between 1993 and 1996.10 We distinguish between
states that prohibited health as a rating factor with
little or no restriction on the use of age (high and low
premiums based on age were allowed to differ by
more than 100%, or a two-to-one band), and those
that prohibited health and imposed a prohibition or
tight restriction on the use of age as well.
368
To control for temporal effects, we compare outcomes for small employers in the nine reform states
with outcomes in a group of 11 states and the District
of Columbia where the reforms were not enacted.11
These states are Alabama, Arkansas, Georgia, Illinois, Indiana, Louisiana, Michigan, Pennsylvania,
New Mexico, Nevada, and West Virginia.12 These 12
areas did not provide for guaranteed issue, nor did
they prohibit health rating prior to 1997.13
Table 1 summarizes the provisions in the nine reform states, indicates the dates of reform, and presents the sample sizes for our study in each state and
the comparison group. We deÀne small employers to
be those with 50 or fewer workers, because this is
the deÀnition adopted in eight of the nine states.
We measure the effect of the regulations in three
ways. First, we compare outcomes in the small group
market in 1997 in each of the reform states with outcomes in the 12 nonreform areas taken as a group.
However, these cross-section comparisons are biased
by self-selection if there are market characteristics
that affect states’ decisions to enact regulations and
affect market outcomes. To control for such statespeciÀc effects, we examine changes in outcomes between the pre-reform period (1993) and the post-reform period (1997), and test whether changes over
the period in each of the reform states differ from
those in the control group. The latter change represents a control for temporal factors. This is typically
referred to as a ‘‘difference-in-differences’’ estimate.
The difference-in-differences estimate accounts for
time-invariant, state-speciÀc effects; however, there
may be state-speciÀc changes over time unrelated to
regulations that could bias this comparison. To control
for state-speciÀc temporal effects, we assume that
these factors affect medium as well as small employers
in the state. (We deÀne medium employers to be those
with 51 to 150 employees.) We then can compare the
change in outcomes over time between small employers and medium employers in reform states with the
change between small and medium employers in the
control states. This is referred to as a ‘‘difference-indifference-in-differences’’ estimate.
Formally, we Àt the following model:
Y ⫽ b0 ⫹ b1 ·YR97 ⫹ b2 ·SMALL ⫹ b3 ·STATE
⫹ b4 ·(STATE·YR97) ⫹ b5 ·(SMALL·STATE)
⫹ b6 ·(SMALL·YR97)
⫹ b 7 ·(SMALL·STATE·YR97) ⫹ b8 ·X
⫹ ⑀.
(1)
Table 1. Small group reform provisions and sample sizes, 1993 and 1997
12 Nonreform
statesa
Reform provisions
1993 provisions
Guarantee issue
Rating: health variation
Rating: age variation
restricted to 100% spread
1997 provisions
Guarantee issue
Rating: health variation
Rating: age variation
restricted to 100% spread
Date of guarantee issue
legislation
Date of rating legislation
Largest size group affected
MD
None
Allowed
None
Allowed
None
Allowed
All
Prohibited
Restricted
NJ
NY
WA
CA
MN
FL
OR
CT
None
None
None
None
Some
Some
Some
None
Allowed
Allowed
Allowed
Allowed
Allowed
Allowed
Allowed
Allowed
Not restricted Not restricted Not restricted Not restricted Not restricted Not restricted Not restricted Not restricted Not restricted Not restricted
All
Prohibited
Restricted
All
Prohibited
Prohibited
Not restricted
NA
Jul-94
Jan-94
Apr-93
Not restricted Not restricted Not restricted Not restricted
Jul-94
Jul-93
Jul-93
Apr-94
Oct-96
Not restricted
NA
NA
NA
Jul-95b
50
Jan-96
49c
Apr-93
50
Jan-96
50
1993 sample sizes
Groups size 1–50
Groups size 51–150
4,436
1,028
375
85
434
100
571
108
426
97
624
115
1997 sample sizes
Groups size 1–50
Groups size 51–150
2,188
440
295
568
97
1,336
193
828
110
1,433
170
d
All
Prohibited
All
Allowed
NA
50
All
Allowed
NA
50
All
Prohibited
All
Prohibited
Restricted
Some
Prohibited
Jan-94
50
Mar-93
25
Jul-95
50
412
114
473
101
368
89
346
76
1,435
134
1,422
141
1,461
166
1,253
112
369
‘‘Second Generation’’ Reforms
Source: National Employer Health Insurance Survey, 1997 Robert Wood Johnson Foundation Employer Health Insurance Survey, and Institute for Health Policy Solutions database of small
group reform provisions.
Note: NA ⫽ not applicable.
a
Alabama, Arkansas, District of Columbia, Georgia, Illinois, Indiana, Louisiana, Michigan, Pennsylvania, New Mexico, Nevada, West Virginia.
b
Maryland enacted prohibitions on health as a rating factor, but with limited restrictions on age effective July 1994. Tighter restrictions on age variation became effective on the date listed.
c
Increased to 50 in October 1997.
d
Fewer than 50 observations.
Inquiry/Volume 38, Winter 2001/2002
The variable YR97 is an indicator that takes the value
1 for observations from the 1997 RWJF survey and
0 for those from the NEHIS. SMALL is an indicator
that takes the value 1 for businesses that are parts of
Àrms with 50 or fewer employees and 0 for businesses of Àrms that have 51 to 150 employees.
STATE represents nine indicator variables included
in the regression for the reform states. For example,
there is one variable that takes the value 1 for employers in Maryland and 0 otherwise; similar variables are included for employers in each of the other
reform states. The vector X represents other characteristics of the employer included to control for differences over time and across states in the composition of employers. We deÀne these control variables
subsequently. The cross-section estimate of the effect
of reform is given by b2 ⫹ b4 ⫹ b6 ⫹ b7; the difference-in-differences estimate is b6 ⫹ b7; and the difference-in-difference-in-differences estimate is b7.
We Àt a model of this form for each of the outcomes
described later.
We use the Àtted model to predict outcomes for
each reform state in each year and for the grouped
control states in each year assuming that the employer composition is the same in all states and over
time.14 We do this by predicting outcomes for each
employer in our RWJF study sample as if it were
observed in the state and year in question. That is,
our RWJF sample is used to deÀne the distribution
of characteristics, X, for the small- and medium-size
employers in our prediction sample. For each of these
employers, we use the Àtted equation given earlier to
predict what the outcome would be in 1993 if the
employer were in one of the nonreform states, by
setting the value for all the nonreform state indicators
to 0 and for YR97 to 0. We then set the YR97 indicator to 1 and predict the 1997 outcome for employers in the nonreform states. Similarly, by varying
the values of the indicators for the reform states we
predict the 1993 and 1997 outcomes for a constant
mix of employers in each of these states. We report
the mean of the predicted values for small employers
for each year and state in the tables to follow. The
difference-in-differences and difference-in-difference-in-differences measures based on the predicted
values are also given. Predicted values for medium
size employers, used in the difference-in-differencein-differences, are reported in the Appendix Table.
Control Measures
In addition to the state, year, and size indicators, our
regression models include other employer character370
istics to control for differences among states in the
composition of employers and employees. These
characteristics include: industry; detailed indicators
for the Àrm and establishment size; the percentage of
low-income workers at the establishment (workers
earning less than $7 per hour); the percentage of
union employees; the percentage of employees working full time (35 hours or more per week); and the
age of the business. We also include indicators for
region in our analysis, so that our comparison between a reform state and the pool of contrast states
is not confounded by regional differences.
Outcome Measures
The outcomes that we examine include insurance offer
and enrollment rates in any employer-sponsored health
plan and in a health maintenance organization (HMO)
plan; turnover in offer decisions; and premiums, variability in premiums, and the change in premiums.
We examine both the percentage of employers offering health insurance and the percentage of employees in Àrms offering insurance. The variable in
each case is the dichotomous response to the question
posed to employers as to whether they offer health
insurance to employees. When we use establishment
weights in the analysis, the result is an estimate of
the percentage of employers offering insurance. Using employee weights, we produce an estimate of the
percentage of employees in businesses that offer insurance. Because the outcome is dichotomous, we Àt
logistic regression models using the variables shown
in equation 1. We also investigate the percentage of
all employees who are enrolled in their employer’s
health insurance plan. This is a continuous variable
and we Àt a linear regression model as shown in
equation 1 for analysis of this outcome.
Some analysts believe that small group insurance
reforms may increase the role of HMOs in the small
group market. Many HMOs use community rating and
do not limit enrollment on the basis of health factors.
Reforms that prohibit these practices by commercial
insurers may, therefore, make the small group market
more attractive to HMOs. Some support for the hypothesis has been demonstrated in other studies (Buchmueller and Jensen 1997; Morrisey and Jensen 1997).
To explore this hypothesis, we Àt a logistic regression
model to examine the percentage of insurance-offering
employers that offer an HMO as one of the plans and
the percentage of enrollees who are in an HMO plan.
Each health insurance plan offered was classiÀed as
an HMO or other type of plan based on the survey
‘‘Second Generation’’ Reforms
respondent’s self-assessment of the plan type, aided by
complete deÀnitions as needed.
Rating reform is intended to lower the price of
insurance for high-risk groups, but, as we noted earlier, the effect on average premiums depends on
whether the reform results in cross-subsidies that
drive low-risk groups from the market (Curtis et al.
1999; Thorpe 1992). The two employer surveys measured premiums for single coverage and for family
coverage for each health insurance plan offered by
employers in the sample. To account for differences
in the scope and depth of beneÀts of different insurance plans, the premiums for each plan were adjusted
by an estimate of the actuarial value of the plan beneÀts. The actuarial value for a plan ranges between
0 and 1 and represents the share of expenditures on
health care by a standardized group of enrollees that
the plan would reimburse. Using the 1987 National
Medical Expenditure Survey, we generated the distribution of expenditures for a representative crosssection of the population under age 65. The 1987
expenditures were adjusted to total spending for 1993
and 1997 using the aggregate spending in each year
from the National Health Accounts, developed by the
Health Care Financing Administration. We then used
information about plan beneÀts from the employer
surveys to determine the amount that each plan
would reimburse.15 We adjusted premiums by multiplying them by the ratio of the average actuarial value for all plans to the plan-speciÀc actuarial value.
We regressed adjusted premiums on the state, year,
and size indicators and the control characteristics described earlier to investigate the effect of the market
reforms on premium levels. We added type of plan
(HMO, preferred provider organization or point-ofservice, indemnity) as a control variable for the premium regressions. The unit of analysis in the premium regressions is the insurance plan. In Àtting the
regression, we used establishment sampling weights
multiplied by the share of the employer’s total enrollment in the plan. For businesses offering more
than one plan, therefore, this effectively examines the
average single premium and the average family premium for policies purchased by the employer.16
Rating reforms restrict the range of variation in
premiums that an insurer is permitted to charge in
the small group market. Thus, they were designed to
reduce variability in premiums among employers and
to promote greater stability over time in premiums
faced by an employer. To examine the effects of reforms on variability in premiums over time, our outcome measure is the residual variance in plan pre-
miums after accounting for the variance attributable
to the control factors (regression variance). That is,
it is the residual variance from the premium regression, measured as the square of the regression residual, for each establishment’s insurance plan. We divided the residual variance in premiums by the
squared value of the mean premium for employers in
the state, year, and size group. That is, our measure
is essentially the coefÀcient of variation squared
(called the relvariance) (Kish 1967). It is a measure
of variance relative to the level of prices to account
for differences in prices over time and geography that
may affect the absolute value of the variance.17
To assess the effects on stability in premiums over
time, we study the percentage of employers reporting
premium changes of greater than 10% (either increase or decrease) between 1996 and 1997 in the
RWJF survey. Absent restrictions on insurer pricing,
small employers can experience extreme rate variation which, in turn, could result in high turnover of
coverage (IHPS 1995). If reforms are successful in
limiting the year-to-year variation in premiums that
small employers face, then there may be greater stability in employers’ decisions to offer insurance
(Buchanan and Marquis 1999). Employers interviewed in the 1997 RWJF survey were asked if they
offered health insurance two years previously as well
as currently. We used these questions to measure
whether the employer made a change in whether to
offer coverage. This variable is used to assess whether rating reform affects stability in the market. Because this measure comes from one point in time, our
model does not include year indicators and the interaction of year and other characteristics.
Limitations
Our data come from two very large surveys and so
provide an ample number of observations to detect
effects of reform, if they exist. Nonetheless, there are
some limitations of our data that may hamper our
ability to detect effects. First, our study period may
be too short to detect the full effects of small group
reform, because it may take some time for insurers
and employers to adjust to the regulations once they
are implemented. Second, we study a period during
which there was considerable activity in states to introduce reforms in the small group market. As a result, our control group was limited to a dozen states,
omitting a broad middle group from the analysis.
Finally, there is the potential for selection bias in
our comparison. Many of the states that we did not
study adopted some regulations on rating and guar371
12 Nonreform
states
MD
NJ
NY
WA
CA
MN
FL
OR
CT
33.5
38.7
⫺2.2
6.6*
7.6
37.9
32.6
⫺8.3*
⫺3.9
⫺.8
48.2
41.5
.6
⫺5.3
⫺5.9
39.7
36.9
⫺4.0
⫺1.4
⫺.5
39.6
44.2
3.3
6.0
10.1
40.5
44.1
3.2
5.0
3.4
45.3
48.4
7.5*
4.5
8.4
42.3
39.2
⫺1.7
⫺1.7
5.7
43.4
46.6
⫺11.1*
3.8
23.7*
50.3
43.8
⫺13.9*
⫺5.9*
⫺18.1
54.6
60.0
2.3
6.0
7.1
57.1
55.4
⫺2.3
⫺1.1
⫺2.5
57.2
59.4
1.7
2.8
6.0
53.9
59.5
1.8
6.2*
8.9
62.7
66.3
8.6*
4.2
4.8
54.8
50.9
⫺6.8*
⫺3.3
7.7
34.7
38.7
⫺.3
3.0
2.1
37.1
33.6
⫺5.4*
⫺4.5*
⫺20.8*
45.4
43.3
4.3
⫺3.1
⫺8.0
38.4
37.3
⫺1.7
⫺2.1
⫺1.4
34.9
39.0
0.0
3.1
1.0
33.5
40.2
1.2
5.7*
1.0
42.5
47.1
8.1*
3.6
⫺3.5
39.4
38.9
⫺.1
⫺1.5
⫺4.1
Percentage of employers offering insurance
1993
1997
1997 difference
Difference in differences
Difference in difference in differencesa
42.3
40.9
NA
NA
NA
43.6
45.9
5.0
3.7
Percentage of employees in Àrms offering insurance
1993
1997
1997 difference
Difference in differences
Difference in difference in differencesa
58.3
57.7
NA
NA
NA
60.8
60.1
2.4
⫺.1
38.0
39.0
NA
NA
NA
36.5
36.2
⫺2.8
⫺1.3
b
b
Percentage of employees enrolled
1993
1997
1997 difference
Difference in differences
Difference in difference in differencesa
b
Source: National Employer Health Insurance Survey, 1997 Robert Wood Johnson Foundation Employer Health Insurance Survey, and Institute for Health Policy Solutions database of small
group reform provisions.
Notes: ‘‘1997 difference’’ compares outcomes in each reform state to those in the 12 nonreform states. ‘‘Difference in differences’’ compares the 1993 to 1997 change in outcomes in each
reform state to the change in the 12 nonreform states. ‘‘Difference in difference in differences’’ compares difference-in-differences measures for employers with 50 or fewer workers to those
with 51 to 150 workers. These measures are not applicable, ‘‘NA,’’ to the nonreform state control group.
a
Predicted values for 1993 and 1997 for employers with 51 to 150 workers are given in the Appendix.
b
Fewer than 50 observations.
* SigniÀcantly different at .05 level.
Inquiry/Volume 38, Winter 2001/2002
372
Table 2. Insurance offer and enrollment rates for small business, 1993 and 1997
Source: 1997 Robert Wood Johnson Foundation Employer Health Insurance Survey and Institute for Health Policy Solutions database of small group reform provisions.
Notes: ‘‘Difference’’ compares turnover in each reform state to that in the 12 nonreform states. ‘‘Difference vs. difference for medium size groups’’ compares the difference for employers
with 50 or fewer workers to that for employers with 51 to 150 workers.
a
Predicted values for 1993 and 1997 for employers with 51 to 150 workers are given in the Appendix.
b
Fewer than 50 observations.
* SigniÀcantly different at .05 level.
18.0
3.3
5.0
8.5
⫺6.2*
⫺.8
16.6
1.9
⫺3.7
b
Percentage of employers changing offer decision 1995 to 1997
Difference
Difference vs. difference for medium size groups
14.7
NA
NA
16.8
2.1
15.7
1.0
3.1
17.0
2.3
3.5
10.5
⫺4.2
⫺.1
9.3
⫺5.4
⫺2.0
16.5
1.8
⫺2.4
OR
WA
NY
NJ
MD
12 Nonreform
states
Table 3. Turnover in offer rates for small business by state reform provisions, 1995 and 1997
CA
MN
FL
CT
‘‘Second Generation’’ Reforms
anteed issue during the study period, though not as
stringent as the reforms we investigated. Some of our
study states adopted less stringent reforms as interim
steps toward the reforms we studied. States that introduced the ‘‘second generation’’ reforms may be
those that found earlier reforms did no harm; in contrast, states that did not move forward may have had
less positive experience with interim reforms. That
is, if reforms were adopted by all states, the effects
might have differed from those we observed.
Results
Our evidence from the employer surveys suggests
that the reforms did not affect coverage (Table 2).
There are few statistically signiÀcant Àndings. Although the cross-section estimate, the difference-indifferences estimate, and the difference-in-differencein-differences estimate often suggest the same direction of effect for a state, there is not a consistent
direction of effect across the states.
Four of the reform states introduced guaranteed
issue for all products and rating reforms (Maryland,
New Jersey, New York, and Washington). The crosssection estimates of the effect of reform are mostly
statistically insigniÀcant; they suggest positive effects
in two states, but negative effects in the other two
states. Similarly, the difference-in-differences estimate and the difference-in-difference-in-differences
estimates are generally statistically insigniÀcant and
yield mixed signs for these four states.18 Two of the
states introduced access reforms but did not prohibit
health as a rating factor (California and Minnesota).
There are no statistically signiÀcant differences between these states and the control states in our three
estimates for the percentage of employers offering
insurance. Furthermore, the direction of effect is positive in one state and negative in the other. Finally,
the other three states (Florida, Oregon, and Connecticut) introduced health rating prohibitions and, in two
of the states, strengthened access reforms as well.
Again, comparison with the control states suggests
statistically insigniÀcant differences, and the direction of effects is of mixed sign. Similar patterns are
found when we examine the percentage of employees
offered insurance and the percentage of employees
enrolled in insurance.19
Previous research has suggested that the effects of
small group reform might differ for the smallest of
small businesses.20 To test for this, we added to our
model of offer rates interaction terms for Àrms with
fewer than 10 workers. For each state, the direction
of effect was similar for very small (⬍10 workers)
373
12 Nonreform
states
Percentage of employers offering HMO (if offer insurance)
1993
1997
1997 difference
Difference in differences
Difference in difference in differencesa
Percentage of enrollees in HMOs
1993
1997
1997 difference
Difference in differences
Difference in difference in differencesa
MD
NJ
NY
WA
CA
MN
FL
OR
CT
14.0
30.0
NA
NA
NA
37.0
58.5
28.5*
5.5
5.6
35.8
5.8
14.2*
29.8*
21.0
43.7
13.7*
6.7
5.1
21.1
33.9
3.9
⫺3.2
7.2
44.7
57.2
27.2*
⫺3.5
⫺8.5
23.5
46.1
16.1*
6.6
18.1
21.4
53.0
23.0*
15.6*
20.8
40.0
57.6
27.6*
1.6
4.2
13.9
37.2
7.2
7.3
24.2
15.2
28.9
NA
NA
NA
32.9
56.2
27.3*
9.6
8.6
33.8
4.9
11.5*
15.5
18.6
43.2
14.3*
10.9
5.5
18.6
22.5
⫺6.4
⫺4.8
⫺13.5
43.5
53.6
24.7*
⫺3.6
13.0
31.1
52.9
24.0*
8.1
⫺.6
25.6
56.2
27.3*
16.9*
29.6*
37.1
55.5
26.6*
4.7
16.1
13.7
36.4
7.5
9.0
33.9*
b
b
Source: National Employer Health Insurance Survey, 1997 Robert Wood Johnson Foundation Employer Health Insurance Survey, and Institute for Health Policy Solutions database of small
group reform provisions.
Notes: ‘‘1997 difference’’ compares outcomes in each reform state to those in the 12 nonreform states. ‘‘Difference in differences’’ compares the 1993 to 1997 change in outcomes in each
reform state to the change in the 12 nonreform states. ‘‘Difference in difference in differences’’ compares difference-in-differences measures for employers with 50 or fewer workers to those
with 51 to 150 workers. These measures are not applicable, ‘‘NA,’’ to the nonreform state control group.
a
Predicted values for 1993 and 1997 for employers with 51 to 150 workers are given in the Appendix.
b
Fewer than 50 observations.
* SigniÀcantly different at .05 level.
Inquiry/Volume 38, Winter 2001/2002
374
Table 4. Offer and enrollment rates for HMOs for small business, 1993 and 1997
Table 5. Monthly premiums and variation in premiums for small business, adjusted for plan beneÀts, 1993 and 1997
12 Nonreform
states
MD
Premiums for employee-only coverage ($)
1993
1997
1997 difference
Difference in differences
Difference in difference in differencesa
159
172
NA
NA
NA
170
166
⫺6
⫺17
Premiums for family coverage ($)
1993
1997
1997 difference
Difference in differences
Difference in difference in differencesa
392
404
NA
NA
NA
430
428
24*
⫺14
c
c
Relvariance in premiums for employee-only coverage
1993
1997
1997 difference
Difference in differences
Difference in difference in differencesa
20.7
17.4
NA
NA
NA
15.7
8.7
⫺8.7*
⫺3.7
Relvariance in premiums for family coverage
1993
1997
1997 difference
Difference in differences
Difference in difference in differencesa
11.2
12.9
NA
NA
NA
11.0
6.5
⫺6.4*
6.2*
NJ
NY
WA
CA
MN
FL
OR
CT
192
205
33*
0
27
196
201
29
⫺8
5
144
158
⫺14
1
8
169
170
⫺2
⫺12
17
138
150
⫺22*
⫺1
14
161
170
⫺2
⫺4
⫺3
130
156
⫺16*
13
21
209
214
42*
⫺8
12
452
474
70*
10
27
446
432
28*
⫺26*
⫺28
387
384
⫺20
⫺15
⫺13
416
413
9
⫺15
⫺5
376
380
⫺24
⫺8
⫺14
406
436
32*
⫺18
⫺14
341
384
⫺20
31*
11
485
503
99*
6
⫺2
30.0
22.0
4.6
⫺4.7
⫺3.5
19.3
16.1
⫺1.3
.1
2.0
16.8
15.0
⫺2.4
1.5
4.1
18.7
13.5
⫺3.9
⫺1.9
⫺5.1
17.9
13.5
⫺3.9
⫺1.1
7.3
16.9
18.8
1.4
5.2
9.7
20.9
22.5
5.1
4.9
4.8
14.6
16.8
3.9
.5
⫺1.3
17.9
16.8
3.9
⫺2.8
⫺3.5
8.1
9.2
⫺3.7
⫺.6
⫺3.2
11.7
11.7
⫺1.2
⫺1.7
⫺5.8
9.6
8.4
⫺4.5*
⫺2.9
⫺1.7
7.3
9.5
⫺3.4
.5
⫺2.4
15.6
14.0
1.1
3.3
⫺5.2
b
c
19.1
25.8
8.4*
10.0*
16.6
b
c
9.8
13.3
.4
1.8
⫺6.0
375
‘‘Second Generation’’ Reforms
Source: National Employer Health Insurance Survey, 1997 Robert Wood Johnson Foundation Employer Health Insurance Survey, and Institute for Health Policy Solutions database of small
group reform provisions.
Notes: Observed premiums have been adjusted to account for differences in plan beneÀts (see ‘‘Outcome Measures’’ in text). ‘‘1997 difference’’ compares outcomes in each reform state to
those in the 12 nonreform states. ‘‘Difference in differences’’ compares the 1993 to 1997 change in outcomes in each reform state to the change in the 12 nonreform states. ‘‘Difference in
difference in differences’’ compares difference-in-differences measures for employers with 50 or fewer workers to those with 51 to 150 workers. These measures are not applicable, ‘‘NA,’’
to the nonreform state control group.
a
Predicted values for 1993 and 1997 for employers with 51 to 150 workers are given in the Appendix.
b
Residual variance from premium level regression divided by the square of the mean premium, times 100.
c
Fewer than 50 observations.
* SigniÀcantly different at .05 level.
376
Source: 1997 Robert Wood Johnson Foundation Employer Health Insurance Survey and Institute for Health Policy Solutions database of small group reform provisions.
Notes: ‘‘Difference’’ compares the rate of large premium change in each reform state to that in the 12 nonreform states. ‘‘Difference in differences’’ compares the difference for Àrms with
50 or fewer workers to that for Àrms with 51 to 150 workers.
a
Fewer than 50 observations.
* SigniÀcantly different at .05 level.
16.4
.4
⫺1.7
18.9
2.4
7.0
28.7
12.2*
7.2
19.8
3.3
1.8
13.0
⫺3.5
.8
22.4
5.9
⫺9.1
8.8
⫺7.7
⫺12.3*
18.2
1.7
3.5
a
OR
FL
MN
CA
WA
NY
NJ
MD
15.3
⫺1.2
16.5
NA
NA
Percentage of employers
Difference
Difference in differences
12 Nonreform
states
Table 6. Small Àrms with premium changes of more than 10% between 1996 and 1997
or other small (10–50 workers) employers (not
shown). However, again there are few statistically
signiÀcant Àndings and no consistent direction of effect across the states.
We also do not Ànd evidence that reforms have
reduced turnover in offers among employers in the
small group insurance market (Table 3). In cross-section, eight of the nine differences are not statistically
signiÀcant; moreover, the point estimates would suggest turnover is greater (not less) in six of the reform
states. Using trends among medium employers as a
control, we also Ànd insigniÀcant differences between the reform states and the control states, and
the results are of mixed sign.21
HMO penetration in the small group market was
generally greater in the reform states than in the nonreform states, both in 1993 and in 1997 (Table 4).
The 1997 difference estimates for HMO offers are all
positive and many are signiÀcant. Overall, the likelihood that an employer offered an HMO in 1997 was
about 18 percentage points higher in the reform states
than the nonreform states; enrollments in the HMOs
in reform states were also about 18 percentage points
higher. These differences are statistically signiÀcant,
but this could be attributed as easily to the higher
initial offer rates as it could to reform. The difference-in-differences estimate and the difference-indifference-in-differences estimate for HMO market
penetration are positive in most reform states, although few are statistically signiÀcant. Averaged
over all the reform states these are also positive, but
not statistically signiÀcant. At best, this offers modest
support for the hypothesis that reforms advanced opportunities for HMOs in the small group market.
The reforms do not appear to have had a substantial effect either on the level or on the variability of
premiums adjusted for plan beneÀts in the small
group market (Table 5).22 In cross-section, we Ànd
higher premiums in some reform states than in the
control states, but lower premiums in others. This is
a weak test, however, because the 1997 differences
primarily reÁect geographic price differences that are
also observed in 1993. The stronger tests, the difference-in-differences estimate and the difference-indifference-in-differences measure, are generally insigniÀcant and of mixed signs.23 Moreover, there is
not evidence of less variability in premiums among
small employers in reform states. In fact, in most
measures, we obtain almost as many positive point
estimates as negative ones and very few of the estimates are statistically signiÀcant. Similarly, we do
not Ànd fewer small employers who faced large pre-
CT
Inquiry/Volume 38, Winter 2001/2002
‘‘Second Generation’’ Reforms
mium changes in either direction in reform states (Table 6). Very few of the differences between reform
and control states are statistically signiÀcant, and
point estimates are of mixed sign.
Buchmueller and Jensen (1997) found that the upper tail of the premium distribution changed more than
the lower tail after reform in California. We tested for
this effect by measuring change in the percentage of
premiums that exceeded 150% of the mean premium,
both for employee-only and family coverage. There
was no support in our data for this hypothesis. The
1997 difference measure for the density of the upper
tail in the reform states in 1997 was positive 10 times
and negative eight times across both types of premiums. Both the difference-in-differences measure and
the difference-in-difference-in-differences measure
were of mixed signs, all were insigniÀcant, and the
measures were often inconsistent within a state for the
two premium types (not shown).
Discussion
Our study adds to the recent literature on the effects
of small group health insurance reform in four ways.
First, it addresses the second generation of reforms,
which were implemented more recently and were
more stringent than the original reforms that several
previous studies have addressed. Second, in choosing
our study states we relied on measures of rate regulation that are more precise than those used by other
researchers, who have used general secondary sources.24 Third, by limiting our comparisons to states with
the most stringent reform provisions and those that
had no reform, our tests were designed to Ànd an
effect, if there is one.25 Finally, our employer survey
data provided large samples of small businesses and
their health insurance offerings, and they permit us
to focus our analysis on the population to which the
reforms were targeted.26 This provided us with a
more convincing test of causation than if we had
studied all employers or employees, or included other
forms of private insurance in addition to employersponsored insurance.
Overall, looking across the several outcomes and
alternative estimators for each of them, we found little evidence that even stringent small group reform
legislation led to a pattern of substantial, signiÀcant
quantitative effects on the small group market across
the nine study states by 1997. Despite the notable
differences in approach and time period, these Àndings are consistent with those of three other recently
published studies (Jensen and Morrisey 1999; Sloan
and Conover 1998; Zuckerman and Rajan 1999).
Both proponents and opponents of small group
market reform will probably greet this news with
mixed emotions. For proponents, we do not Ànd support for the hope that reform would lead to a significant narrowing of the variance in premiums or a
sizable expansion in employment-based coverage
among small groups. On the other hand, there might
have been important improvements in access for
some high-risk groups, which we could not detect
with our analysis. For opponents of reforms of small
group insurance markets, there should be some relief
in hearing that stringent reform did not lead to marketwide price increases or sizable reductions in employment-based coverage. On the other hand, our
analysis would not detect whether or not some lowrisk groups might have dropped coverage in response
to speciÀc rate increases they faced.
The reforms that we studied are in large part intended to eliminate some of the most serious problems in the small group market. They sought to guarantee that no small business is denied access to insurance, to protect groups that experience unexpected
high medical expenditures from extreme premium
volatility, and to shield high-risk groups from prohibitively high costs of insurance. However, our analysis is targeted toward Ànding overall effects across
the whole population of small employers. This does
not mean that there were not many individual gainers
and losers under reform in the various states we studied. However, identiÀcation of these potential effects
will require other designs and data sources.
Notes
Any views expressed in this paper are solely those of the
authors, and no endorsement by the Robert Wood Johnson
Foundation, the National Center for Health Statistics, or
RAND is intended or should be inferred. The authors are
grateful to Rick Curtis, Rafe Forland, Kevin Haugh, and
Stephanie Lewis for coding and interpreting the small
group reform data used to classify states in this research.
The authors thank Linda Andrews, Roald Euller, and Ellen
Harrison for their efforts in preparing the survey data Àles
on which this paper is based.
1 Sloan and Conover (1998) studied nonelderly adults only
during the period 1989–94. Zuckerman and Rajan (1999)
studied all nonelderly during the period 1989–95.
2 This study used an augmented version of Zuckerman
and Rajan’s data, for the same population and period,
but it coded small group reforms differently.
377
Inquiry/Volume 38, Winter 2001/2002
3 Simon reports a signiÀcant but small (2 percentage
point) fall in offer rates among small employers in
states implementing reform relative to the trend in nonreform states. After adjusting for temporal change
among large employers, however, the effect is not signiÀcant.
4 This database was developed by the Institute for Health
Policy Solutions, see Curtis et al. (1999).
5 For details on the NEHIS, see National Center for
Health Statistics (1997).
6 See Research Triangle Institute (1998); Kemper et al.
(1996).
7 Excluded are self-employed individuals with no employees.
8 Prior to 1993, only two states, Vermont and Massachusetts, had legislation which guaranteed issue of all products in the small group market and prohibited health as
a rating factor. This and other results on the history of
state reform legislation are from an analysis of the IHPS
database documenting the legislative history of reforms
in all states (Curtis et al. 1999).
9 The reforms in one of our states—New York—were
effective in April 1993. However, some policies had
until their Àrst anniversary following the effective date
to comply, and so we assume reforms had little effect
in 1993.
10 California and Minnesota also introduced limits on the
use of health status as a rating factor, but did not prohibit it. In California, the high and low premium based
on health cannot differ by more than 20%; in Minnesota, the difference is limited to 67%.
11 Two of these 12 states, Michigan and Pennsylvania, had
Blue Cross–Blue Shield plans that had signiÀcant market share and practiced guaranteed issue and community
rating. We tested the sensitivity of the results shown in
this paper by excluding these two populous states from
the control. Our Àndings were the same and so are not
shown here.
12 The states not included in our analysis enacted some
new regulations concerning the issue of policies or rating reforms but did not adopt regulations as stringent
as the reform states. However, because these states did
undergo regulatory change that might affect outcomes,
their inclusion could dampen our estimate of the effect
of the reforms we study. Our objective was to maximize
our ability to detect effects.
13 By mid-1997, most states had enacted legislation to
guarantee issue of all products in the small group market to conform to the HIPAA legislation. We assume
that this does not affect outcomes for 1997, because
most contracts are written for the calendar year. None
of our nonreform states adopted guarantee issue provisions for even a limited range of products prior to this.
Some of our comparison states had rating reforms, however, but they permitted considerable variation in premiums based on age and health status (greater than
100% spread).
14 We report results for each of the reform states separately because we want to investigate whether there is
a consistent pattern of effects across the states. Moreover, the states differed in the nature of reforms enacted (see Table 1). In contrast, we pool the 12 nonreform states. While the magnitude of a reform effect
might be sensitive to the choice of a contrast, our conclusions stem in large part from a lack of consistency
across the reform states—and different choices of the
nonreform states to include in the contrast would not
alter this Ànding. Furthermore, the contrast states are
alike in that they did not enact reforms on the dimen-
378
15
16
17
18
19
20
21
22
23
24
sions we study. Moreover, as we noted earlier, the exclusion of two nonreform states in the contrast led to
similar results.
The expenditure distributions and actuarial values were
designed and calculated for us by the Actuarial Research Corporation.
We also produced estimates using enrollee weights and
obtained essentially the same results, so they are not
reported.
Using the variance rather than relative variance as our
outcome yielded similar results.
The predicted values on all outcomes for medium-size
employers, which enter into the difference-in-difference-in-differences calculation, are given in the Appendix Table.
To conÀrm our Àndings from the employer survey, we
conducted a similar analysis for the share of workers
enrolled in their own employer group plan using data
from the March 1993 and March 1997 Current Population Survey (CPS). With the CPS, we analyzed the
difference-in-differences between each reform state and
the 12 nonreform states in the percentage of employees
in Àrms with fewer than 100 workers who held coverage from their own employer; the CPS does not permit
us to produce estimates for Àrms of 50 or fewer workers. The results from the CPS also did not produce statistically signiÀcant estimates of the effect of reform,
nor did the results suggest a consistent pattern of effects
across states.
See, for example, Buchmueller and Jensen (1997).
Because we are comparing turnover in cross-section, we
do not have controls for temporal effects. Moreover, the
Àrst test which contrasts stability in each of the reform
states with the nonreform states does not control for
state-speciÀc effects. The second test uses stability of
medium-size employers as control for state-speciÀc effects. These are weaker tests than employed for the other outcome measures. However, we report these results
because improved stability in the small group market
was an important goal of reforms.
Because premiums are highly skewed, we also Àt models using the natural logarithm of premiums. These
models produced few signiÀcant difference-in-differences or difference-in-difference-in-differences estimates of the effect of reform. As with the estimates of
the effects on the mean, the effects on the median were
not of consistent sign. We present the predictions from
the model Àt to premiums on the dollar scale rather than
the logarithm scale, because the former produces estimates of mean differences, whereas we would have to
make assumptions about how variances vary across
states to transform the predictions from the log scale to
the dollar scale.
Simon (1999), in contrast, concludes that states that
have some form of guaranteed issue and some rating
regulations have higher premiums than states without
these regulations, based on an analysis of the NEHIS
and the Medical Expenditure Panel Survey Insurance
Component completed in 1996. However, the trend in
premiums among small employers (those with 25 or
fewer employees) did not differ signiÀcantly between
reform and nonreform states and was very small in
magnitude. The conclusion stems primarily from a negative trend in premiums for large employers (those with
100 or more employees) in the reform states relative to
the nonreform states (that is, from a difference-in-difference-in-differences estimate of the effect).
This is a step for further research urged by Marsteller
et al. (1998). In addition, our choices of states were
‘‘Second Generation’’ Reforms
based on implementation dates for both rate regulation
and guaranteed issue that are more precise than the assumptions that have been used before.
25 By considering each reform state separately, in effect
we follow Zuckerman and Rajan (1999), who advised
examining reforms as packages.
26 Jensen and Morrisey (1999) acknowledged the desirability
of large samples of employers for this type of research.
References
Bilheimer, L.T., and D.C. Colby. 2001. Expanding Coverage: ReÁections on Recent Efforts. Health Affairs 20(1):
83–95.
Blumberg, L.J., and L.M. Nichols. 1996. First Do Not
Harm: Developing Health Insurance Market Reform
Packages. Health Affairs 15(3):35–53.
Branscome, J.M., P.F. Cooper, J. Sommers, and J.P. Vistnes.
2000. Private Employer-Sponsored Health Insurance:
New Estimates by State. Health Affairs 19(1):139–147.
Buchanan, J. L., and M. S. Marquis. 1999. Who Gains and
Who Loses with Community Rating for Small Business? Inquiry 36(1):30–43.
Buchmueller, T.C., and G. A. Jensen. 1997. Small Group
Reform in a Competitive Managed Care Market: The
Case of California, 1993 to 1995. Inquiry 34(3):249–
263.
Buchmueller, T., and J. DiNardo. 1998. Did Community
Rating Induce an Adverse Selection Death Spiral? Evidence from New York, Pennsylvania, and Connecticut.
Unpublished. University of California, Irvine.
Chollet, D. J., and R.R. Paul. 1995. Community Rating:
Issues and Experience. Washington, D.C.: Alpha Center.
Curtis, R., S. Lewis, K. Haugh, and R. Forland. 1999.
Health Insurance Reform in the Small-Group Market.
Health Affairs 18(3):151–160.
Custer, W.S. 1998. Health Insurance Coverage and the Uninsured. Washington, D.C.: Health Insurance Association of America.
Gabel, J.R., P.B. Ginsburg, and K.A. Hunt. 1997. Small
Employers and Their Health BeneÀts, 1988–1996: An
Awkward Adolescence. Health Affairs 16(5):103–110.
Herzlinger, R.E. 2000. Leveling the Playing Field. The New
Democrat Blueprint 6:22.
Hing, E., and G.A. Jensen. 1999. Health Insurance Portability and Accountability Act of 1996, Lessons from the
States. Medical Care 37(7):692–705.
Institute for Health Policy Solutions (IHPS). 1995. State
Experiences with Community Rating and Related Reforms. Washington, D.C.: IHPS.
Jensen, G.A., and M.A. Morrisey. 1999. Small Group Reform and Insurance Provision by Small Firms, 1989–
1995. Inquiry 36(2):176–187.
Kemper, P., D. Blumenthal, J. M. Corrigan et al. 1996. The
Design of the Community Tracking Study: A Longitudinal Study of Health System Change and Its Effects
on People. Inquiry 33(2):195–208.
Kish, L. 1967. Survey Sampling. New York: John Wiley
and Sons, Inc.
Marsteller, J.A., L. M. Nichols, A. Badawi, et al. 1998.
Variations in the Uninsured: State and County Level
Analyses. Washington, D.C.: The Urban Institute.
Morrisey, M.A., and G.A. Jensen. 1997. Switching to Managed Care in the Small Employer Market. Inquiry 34(3):
237–248.
National Center for Health Statistics (NCHS). 1997. Employer-Sponsored Health Insurance: State and National
Estimates. Hyattsville, Md.: NCHS.
Research Triangle Institute (RTI). 1998. 1997 Employer
Health Insurance Survey: Final Methodology Report.
Research Triangle Park, N.C.: RTI.
Simon, K.J. 1999. The Impact of Small-Group Health Insurance Reform on the Price and Availability of Health
BeneÀts. Working Paper. University of Maryland.
Sloan, F.A., and C. J. Conover. 1998. Effects of State Reforms on Health Insurance Coverage of Adults. Inquiry
35(3):280–293.
Thorpe, K.E. 1992. Expanding Employment-Based Health
Insurance: Is Small Group Reform the Answer? Inquiry
29(2):128–136.
Wessner, C. 1994 State Community Rating Reforms: Early
Tests Produce Mixed Result. State Health Notes, Intergovernmental Health Policy Project 15(193):1–8.
Zuckerman, S., and S. Rajan. 1999. An Alternative Approach to Measuring the Effects of Insurance Market
Reforms. Inquiry 36(1):44–56.
379
Outcome
12 Nonreform
states
MD
NJ
NY
WA
CA
MN
FL
OR
CT
91.2
90.2
92.7
86.5
84.5
86.6
82.5
94.5
94.1
95.3
93.4
88.2
83.1
77.2
77.8
91.9
87.0
91.6
83.2
Percentage of employees offered insurance
1993
1997
90.5
93.9
93.2
89.7
73.2
77.7
93.3
95.0
97.3
94.1
93.9
88.2
88.4
81.7
82.4
92.2
95.0
85.7
78.1
Percentage of employees enrolled
1993
1997
59.5
56.6
55.9
55.9
53.9
54.5
67.9
67.7
69.7
63.7
60.1
58.2
57.4
51.0
52.8
68.8
73.0
56.1
55.8
5.6
6.5
3.6
4.3
11.9
13.3
2.3
6.0
26.3
25.8
28.1
44.8
30.5
35.2
50.2
70.3
36.7
40.3
40.1
50.0
43.2
55.7
25.6
23.8
18.4
30.8
27.6
49.4
15.8
40.9
57.6
57.4
39.2
64.3
52.8
56.5
37.0
42.0
31.8
23.3
Percentage of employers offering insurance
1993
1997
Percentage of employers changing offer decision 1995 to 1997
7.7
a
a
a
a
Percentage of employers offering HMO (if offer insurance)
1993
1997
24.6
39.7
44.5
Percentage of enrollees in HMOs
1993
1997
23.6
40.0
31.1
a
a
Monthly premiums for employee-only coverage ($)
1993
1997
152
168
155
Monthly premiums for family coverage ($)
1993
1997
388
392
412
Relvariance in premiums for employee-only coverage
1993
1997
Relvariance in premiums for family coverage
1993
1997
Percentage of Àrms with premium change of more 10%, 1996 to 1997
a
a
184
186
184
200
145
167
160
160
133
147
136
164
129
150
182
191
445
432
410
416
404
406
409
403
370
380
370
406
350
374
463
475
11.7
16.6
6.1
.8
15.3
18.7
20.1
23.7
19.0
21.7
22.1
19.0
9.0
15.1
12.8
7.4
15.0
16.5
16.7
21.6
7.4
7.5
5.3
10.9
12.8
12.5
13.3
6.9
9.6
4.1
12.0
4.0
8.2
5.3
4.2
4.4
7.4
11.3
13.3
11.4
17.7
28.2
8.9
14.6
18.2
8.6
13.2
15.3
a
a
Source: National Employer Health Insurance Survey, 1997 Robert Wood Johnson Foundation Employer Health Insurance Survey, and Institute for Health Policy Solutions database of small
group reform provisions.
a
Fewer than 50 observations.
Inquiry/Volume 38, Winter 2001/2002
380
Appendix Table. Predicted 1993 and 1997 outcomes for medium-size employers.